Tensorflow numpy api

[SebastianJL]

Done so far see zfit/zfit#310.

• tensorflow.experimental.numpy.experimental_enable_numpy_behavior(). Namespace may differ.
• The modules z.unstable, z.wrapping_tf and z.zextension now both use tensorflow NumPy and pure tensorflow functions in an inconsistent manner. The purpose of these three modules seems overlapping at times and I think it would make sense to harmonize them at a later point when tensorflow NumPy has proven to be stable enough. I.e. remove parts of it in favour of tensorflow numpy and consolidate the parts that are still necessary. This is also connected to the next point.
• numpy.py expose functions better. At the moment you can neither find the available functions nor a documentation.
  Simple idea: Generate from tf.exp.numpy import sin statically.
• Expose znp to user.
  • Docstrings.
  • Tutorial, 5 minutes to zfit etc.
  • Examples
• Maybe change type hints?